A Jump Di®usion Model for Option Pricing with

Three Properties: Leptokurtic Feature, Volatility

Smile, and Analytical Tractability
S. G. Kou¤
Columbia University

First draft November, 1999

Abstract
Brownian motion and normal distribution have been widely used, for example, in the
Black-Scholes-Merton option pricing framework, to study the return of assets. However,
two puzzles, emerged from many empirical investigations, have got much attention recently,
namely (a) the leptokurtic feature that the return distribution of assets may have a higher
peak and two (asymmetric) heavier tails than those of the normal distribution, and (b) an
empirical abnormity called \volatility smile" in option pricing. To incorporate both the
leptokurtic feature and \volatility smile", this paper proposes, for the purpose of studying
option pricing, a jump di®usion model, in which the price of the underlying asset is modeled
by two parts, a continuous part driven by Brownian motion, and a jump part with the
logarithm of the jump sizes having a double exponential distribution. In addition to the
above two desirable properties, leptokurtic feature and \volatility smile", the model is simple
enough to produce analytical solutions for a variety of option pricing problems, including
options, future options, and interest rate derivatives, such as caps and °oors, in terms of the
Hh function. Although there are many models can incorporate some of the three properties
(the leptokurtic feature, \volatility smile", and analytical tractability), the current model
can incorporate all three under a uni¯ed framework.

1. lntroduction

Brownian motion and normal distribution have been widely used to study option pricing and the
return of assets; for references, see, for example, Cox and Rubinstein (1985), Du±e (1995), Hull
(1999), Ingersoll (1988), Karatzas and Shreve (1998), Merton (1990), Musiela and Rutkowski
(1997), Elliot and Kopp (1998), and Boyle, Broadie, and Glasserman (1997). Option pricing
papers within the classical Black-Scholes-Merton model that are particularly relevant to the
¤
312 Mudd Building, Department of IEOR, Columbia University, New York, NY 10027, e-mail:
kou@ieor.columbia.edu. The mathematica code used in the current paper can be downloaded from the web
page www.ieor.columbia.edu/~kou
current paper are: Black and Scholes (1973) model for the call and put options; Black (1976)
model for options on futures contracts; Heath, Jarrow and Morton (1992) model for options on
bonds; and Brace, Gatarek, and Musiela model (1997) for caps and °oors, which are options
on discretely compounded simple interest rates and are among the most liquated interest rate
options (see also Miltersen, Sandmann and Sondermann, 1997, and Jamshidian, 1997).
Despite the successes of Black-Scholes-Merton model based on Brownian motion and normal
distribution, two puzzles, emerged from many empirical investigations, have got much attention
recently.
(1). The leptokurtic and asymmetric features. In the above classical models, the marginal
distribution of the underlying assets is assumed to be normal. However, many empirical studies
suggest that the distribution is skewed to the left, and has a higher peak and two heavier tails
than those of the normal distribution.
(2). The volatility smile. More precisely, if the Black-Scholes-Merton model is correct, then
the implied volatility should be constant; but it is widely recognized that the implied volatility
curve resembles a \smile" , meaning it is a convex curve of the strike price.
Many researches have been conducted to modify the Black-Scholes models to explain the
two puzzles. To incorporate the leptokurtic and asymmetric features, a variety of models have
been proposed, including, among others, (a) chaos theory, fractal Brownian motion, and stable
processes; see, for example, Mandelbrot (1963, 1967), Mandelbrot, Fisher, and Calvet (1997),
Fama (1963, 1965), Rogers (1997), Willinger, Taqqu, and Teverovsky (1999), Samorodnitsky
and Taqqu (1994), Peters (1991, 1994); (b) generalized hyperbolic models, including log t-model,
log hyperbolic model, and log variance gamma model; see, for example, Madan and Seneta
(1990), Eberlein and Keller (1995), Barndor®-Nielsen (1995), Praetz (1972), Blattberg and
Gonedes (1974); (c) time changed Brownian motions; see, for example, Clark (1973), Andersen
(1996), Hurst, Platen and Rachev (1997), Geman, Madan, and Yor (1998), and Heyde (1999).
An immediate problem with these models is that it may be di±cult to obtain analytical solutions
for the purpose of option pricing; more precisely, they might give some analytical formulae for
regular call and put options, but certainly not for interest rate derivatives and exotic options,
such as perpetual American options, barrier and lookback options.
In a parallel development, di®erent models are also proposed to incorporate the \volatility
smile". Popular ones are (a) stochastic volatility and ARCH models; see, for example, Hull
and White (1987), Engle (1982, 1995), White (1980), Gouri¶eroux (1997); (b) constant elasticity
model (CEV) model; see, for example, Cox and Ross (1976), Cox, Ingersoll and Ross (1985),

2
Davydov and Linetsky (1999), Andersen and Andreasen (1999); (c) normal jump models, ¯rst
proposed by Merton (1976) and widely used since then; see, for example, Merton (1990), and
Du±e (1995); (d) a numerical procedure called \implied binomial trees"; see, for example,
Derman and Kani (1994), Dupire (1994), Rubinstein (1994). Aside from the problem that it
might not be easy to ¯nd analytical solutions for option pricing, especially for exotic options
(such as perpetual American options, barrier and lookback options), these models may not
produce the leptokurtic and asymmetric features, especially the \high peak" feature.
The current paper attempts to propose a new model, which has three properties.

² It has the leptokurtic and asymmetric features, under which the return distribution of
the assets has a higher peak and two heavier tails than the normal distribution, especially
the left tail; see section 2.

² It leads to analytical solutions to many option pricing problems, including

{ call and put options, and options on futures; see section 4.

{ interest rate derivatives, such as caplets, caps, and bond options; see section 5.
{ exotic options, such as perpetual American options, barrier and lookback options,
which will be reported in a separate paper.

² It can reproduce the \volatility smile"; see section 5.2.

Although there are, as we discussed before, many models that can incorporate some of the
three properties (the leptokurtic feature, analytically tractability, and \volatility smile"), the
current model can incorporate all three under a uni¯ed framework.
The model that we propose for the price of an underlying asset (for example, a stock or a
stock index) is very simple. It consists of two parts, a continuous part modeled by a geometric
Brownian motion, and a jump part, with the logarithm of the jump sizes having a double
exponential distribution and the jump times corresponding to the event times of a Poisson
process. Because of the simplicity, the parameters in the model can be easily interpreted, and
the closed form solutions for option pricing can be obtained in terms of the Hh functions.
General properties of jump di®usion models with independent identically distributed jump
sizes have been extensively studied since the original paper of Merton (1976); for excellent
surveys, see Du±e (1995) and Merton (1990). In addition to the modeling and studying of
the leptokurtic feature and \volatility smile", the technical contribution of the current paper is

3
that we provide an explicit calculation of option prices in the case of the logarithm of the jump
sizes being double exponentially distributed. The explicit calculation is made possible partly
because of the memoryless property of the double exponential distribution.
The paper is organized in the following way. In section 2, the model is proposed and the
leptokurtic feature is studied. Some preliminary results, including the Hh functions, are given
in section 3. Formulae for option pricing problems, including options on futures, are provided
in section 4. Section 5.1 studies the pricing of interest rate options, such as caplets and bond
options. \Volatility smiles" is studied in section 5.2. The last section discusses the advantages
and disadvantages of the model.

2. The model

The model that we propose for the price of an underlying asset (for example a stock or a stock
index) consists of two parts, a continuous part modeled by a geometric Brownian motion, and
a jump part, with the logarithm of the jump sizes having a double exponential distribution
and the jump times corresponding to the event times of a Poisson process. More precisely, the
following stochastic di®erential equation is used to model the asset price, S(t),
0 1
N(t)
dS(t) X
= ¹dt + ¾dW (t) + d @ (Vi ¡ 1)A ; (2.1)
S(t) i=1

where W (t) is a standard Wiener process, N(t) a Poisson process with rate ¸, and fVi g a
sequence of independent identically distributed (i.i.d.) nonnegative random variables such that
X = log(V ) has a double exponential distribution with the density
1 ¡jx¡·j=´
fX (x) = e ; 0 < ´ < 1:
2´
In other words, ( )
»; with probability 1/2
X ¡·= ; (2.2)
¡»; with probability 1/2
where » is an exponential random variable with mean ´ and variance ´2 . All sources of ran-
domness, N(t), W (t), and X's, are assumed to be independent.
Remark. For notation simplicity, and in order to get some analytic solutions for various
option pricing problems, here the drift ¹ and the volatility ¾ are assumed to be constants,
and the Wiener processes and jumps are assumed to be one-dimensional. These assumptions,
however, can be easily dropped for the purpose of developing a general theory.

4
 Solving the stochastic di®erent equation (2.1) gives the dynamics of the asset price as follows:
½ ¾ N(t)
Y
1
S(t) = S(0) exp (¹ ¡ ¾2 )t + ¾W (t) Vi : (2.3)
2 i=1

Merton (1976) ¯rst considered the jump di®usion models similar to (2.1) and (2.3). In that paper
X's are assumed to have normal distribution rather than the double exponential distribution,
although some general properties, for the models with arbitrary distributions, were discussed
there. A major goal of this paper is to show that, within this very simple jump di®usion
framework, it is possible to get some desirable features of the return of the asset, such as higher
peak and heavier tails, particularly the left tail, as well as retaining analytic tractability of the
model, so that options can, in terms of the Hh functions, be priced in closed form.
To motive further studies of the model, it would be of interest to discuss the return of the
underlying asset in such a model. Using (2.3), we get
¢S(t) S(t + ¢t)
= ¡1
S(t) S(t)
8 9
< 1 X = N(t+¢t)
= exp (¹ ¡ ¾2 )¢t + ¾(W (t + ¢t) ¡ W (t)) + Xi ¡ 1;
: 2 ;
i=N(t)+1

where a summation over an empty set is taken to be zero. If the time interval ¢t is small, as in
the case of daily observations, the return can be approximated, ignoring the terms with order
higher than ¢t and using the expansion ex ¼ 1 + x + x2 =2, by
N(t+¢t)
¢S(t) 1 X 1
¼ (¹ ¡ ¾ 2 )¢t + ¾(W (t + ¢t) ¡ W (t)) + Xi + ¾2 (W (t + ¢t) ¡ W (t))2
S(t) 2 i=N(t)
2
N(t+¢t)
p X
¼ ¹¢t + ¾Z ¢t + Xi ;
i=N(t)+1

where Z is a standard normal random variable. Notice that the probability of the Poisson
process N(t) having one jump is ¸¢t, and the probability of having more than one jumps is
o(¢t). Therefore, if ¸¢t is small, ignoring multiple jumps leads to
N(t+¢t) ( )
X Xi with prob. ¸¢t
Xi ¼ :
0 with prob. 1 ¡ ¸¢t
i=N(t)+1

In summary, for small ¢t, the return can be approximated, in distribution, by

¢S(t) p
¼ ¹¢t + ¾Z ¢t + B ¢ X; (2.4)
S(t)

5
where B is a Bernoulli random variable with P (B = 1) = ¸¢t and P (B = 0) = 1 ¡ ¸¢t, and
X is given by (2.2). Notice that dropping the last term in (2.4) leads to the classical model of
geometric Brownian motion, with the return, ¢S(t)=S(t), being characterized approximately
by a normal density.
By using the result (3.16), in section 3, of the density of sum of double exponential and the
normal random variable, we have that the density, g, of the right hand side of (2.4), being an
¢S(t)
approximation for the return S(t) , is given by
à !
¸¢t ¾2 ¢t=(2´2 ) ¡(x¡¹¢t¡·)=´ (x ¡ ¹¢t ¡ ·)´ ¡ ¾2 ¢t
g (x) = e fe © p
2´ ¾´ ¢t
à !
(x ¡ ¹¢t ¡ ·)´ + ¾2 ¢t
(x¡¹¢t¡·)=´
+e © ¡ p g
¾´ ¢t
1 x ¡ ¹¢t
+ (1 ¡ ¸¢t) p '( p );
¾ ¢t ¾ ¢t
where '(¢) is the standard normal density function and · = E(X). The density g has the mean
and the variance given by
Eg (G) = ¹¢t + ¸¢t· (2.5)

V arg (G) = ¾2 ¢t + 2´2 ¸¢t + ·2 ¸¢t(1 ¡ ¸¢t): (2.6)

Remark. An important feature of this density is that, comparing to the normal density
with the identical mean and variance, it has a higher peak around the mean, and two heavier
tails, or, in short, the leptokurtic feature. The other thing, worthy of mentioning, is that the
density is not symmetric, if the mean jump size · is not zero; in fact, it is skewed to the left if
· < 0, and skewed to the right if · > 0. These features have been favored by many empirical
investigations.
Below are ¯gures of the density, g(x), compared with the normal density with the same
mean and variance given by (2.5) and (2.6). The ¯rst ¯gure compares the overall shapes of the
two densities, the second one details the shapes around the peak areas, and the last two show
the left and right tails. The dot line is used for the normal density, and the solid line is used
for the model. The parameters used here are ¢t = 1 day = 1=250 year, ¾ = 20% per year,
¹ = 15% per year, ¸ = 10 per year, · = ¡2%; ´ = 2%. In other words, there are about 10
jumps per year with average jump size ¡2%, and the jump volatility 2%. The jump parameters
used here seem to be quite reasonable, if not conservative, for a U. S. stock. The leptokurtic
feature, however, is quite evident. The peak of the density g is about 30.6, whereas that of the

6
normal density is about 27.7. The density g has heavier tails than the normal density, especially
for the left tail, which could reach ¡10% while the normal density is basically con¯ned within
¡4%:
Remark. Additional numerical plots suggest that the feature of higher peak and heavier
tails becomes more signi¯cant if either j·j (the jump size) or ´ (the jump volatility) increases.
Remark. Although it is possible to get heavier tails by using the normal distribution for
the logarithm of the jump sizes, instead of the double exponential distribution, it is impossible
for it to have both high peak and heavier tails.

30 30

25 28

26
20
24
15
22
10 20

5 18

16
-0.1 -0.08 -0.06 -0.04 -0.02 0 0 0.02 0.04 0.06 0.08 0.1 -0.01 -0.008-0.006-0.004-0.002 0 0.002 0.004 0.006 0.008 0.01
x x

Figure 2.1: Overall comparison Figure 2.2: Peak comparison

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

-0.1 -0.08 -0.06 -0.04 -0.02 00 00 0.02 0.04 0.06 0.08 0.1
x x

Figure 2.3: Left tail comparison Figure 2.4: Right tail comparison

3. Some preliminary results

To price options, we have to notice that our jump di®usion model leads to an incomplete market;
therefore, the standard hedging arguments may not be useful here. However, as we mentioned
before, many papers have studied the general properties of various jump di®usion models. In

7
particular, standard results tell us that (see, for example, Du±e 1995, and Merton, 1990), for
a given set of risk premiums, we can consider a risk-neutral measure P¤
0 1
N(t)
dS(t) X
= (r ¡ ¸E(V ¡ 1))dt + ¾dW (t) + d @ (Vi ¡ 1)A
S(t) i=1
0 1
N(t)
X
= (r ¡ ¸³)dt + ¾dW (t) + d @ (Vi ¡ 1)A ;
i=1
e·
where ³ = 1¡´ 2
¡ 1, 0 < ´ < 1, via (3.3); and the parameters; ·; ´, ¸, and ¾, here are no
longer physical parameters, but the risk-neutral parameters taking consideration also of the
risk premiums. The unique strong solution of the above equation is given by
½ ¾ N(t)
Y
1
S(t) = S(0) exp (r ¡ ¾2 ¡ ¸³)t + ¾W (t) Vi :
2 i=1
For pricing of European options in the jump di®usion model, we need to compute the expecta-
tion, under the measure P¤ , of the discounted ¯nal payo® of the option. In particular, the price
of a call option at time 0, Ãc (0), is given by
³ ´
Ãc (0) = E¤ e¡rT Ãc (T )
0 0 (Ã ! ) N(T ) 1+ 1
¾2 p Y
= E¤ @e¡rT @S (0) exp r¡ ¡ ¸³ T + ¾ T Z Vj ¡ K A A ; (3.1)
2 j=1

where Ãc (T ) = (S(T ) ¡ K)+ and Z is a standard normal random variable. Notice the fact,
which will be used later, that under this measure P¤
8 9
³ ´ < ½ ¾ N(t)
Y =
1
E¤ e¡rT S(T ) = e¡rT S(0)E¤ exp (r ¡ ¾2 ¡ ¸³)t + ¾W (t) Vi
: 2 ;
i=1
8 9
< N(t) =
Y
= S(0)E¤ exp f¡¸³tg Vi
: ;
i=1
1
( n
)
X Y (¸t)n ¡¸t
¤
= S(0) exp f¡¸³tg E Vi e
n=0 i=1
n!
X1
(¸t)n ¡¸t
= S(0) exp f¡¸³tg (³ + 1)n e
n=0
n!
¸(³+1)t ¡¸t
= S(0) exp f¡¸³tg e e ;

in other words,
³ ´
E¤ e¡rT S(T ) = S(0); (3.2)
justifying the name \risk-neutral".

8
3.1. Double exponential distribution

The double exponential density, de¯ned by

1 ¡jx¡·j=´
fX (x) = e ; 0 < ´ < 1;
2´
was proposed by Laplace (1774), giving rise to another name | \ the ¯rst law of Laplace",
while the \second law of Laplace" is the normal density. It can also be represented as
( )
» with probability 1=2
X ¡·= ;
¡» with probability 1=2

where » is an exponential random variable with mean ´ and variance ´ 2 : One of the most
important properties of this density is that it has the leptokurtic feature, namely it has a
higher peak and two heavier tails than those of the normal density with the identical mean and
variance; see, for example, Kotz, Johnson, and Balakrishnan (1995).
Remark. A unique feature, inherited from the exponential distribution, about the double
exponential distribution is the memoryless property, to be used in the next subsection. It is
because of this memoryless property, closed form solutions for various option pricing problems
are feasible.
Although, at the point of the mean, the double exponential density has a discontinuity of the
derivative, in our model of the option pricing it will be smoothed out by the normal distribution
from the continuous Brownian motion part, resulting in a density everywhere di®erentiable.
A useful result of the double exponential random variable is that
e·
E(eX ) = ´ ³ + 1; 0 < ´ < 1; (3.3)
1 ¡ ´2
since Z Z Ã !
0 1 x=´ 1 1 ¡x=´ 1 1 1 1
x x
e e dx + e e dx = 1 + 1 = :
¡1 2´ 0 2´ 2´ 1+ ´ ´ ¡1 1 ¡ ´2
Remark. In a forthcoming paper by Heyde and Kou (2000) they point out that, for a
sample size of 5000 (20-years-daily data), it is very di±cult to distinguish the double exponen-
tial distribution from the power type distributions, such as t-distribution; and, in fact, they
also show that many statistical procedures may fail to detect di®erences between the double
exponential distribution and t-distributions. For example, the following histograms suggest
that they are almost indi®erent to human eyes; for details, see Heyde and Kou (2000). It
should be emphasized, however, that in the current paper we are not arguing which one is
more suitable for ¯tting empirical observations, but just to say that, since the t-distribution

9
cannot, in general, give closed form solutions, the double exponential distribution may be a
good alternative choice as a balance between reality and analytical tractability; furthermore,
t-distribution cannot have the high peak feature.

Figure 3.1: Histogram of the double exponential distribution with mean 0 and variance 1.

Figure 3.2: Histogram of the t-distribition with d.f. 6.

3.2. Sum of double exponential random variables

Let Xi , i = 1; 2; :::; be a sequence of independent identically distributed (i.i.d.) double expo-

nential random variables with
( )
d »; with probability 1/2
X ¡·= ;
¡»; with probability 1/2
where » is an exponential random variable with mean ´ and variance ´ 2 . Notice that

E(X) = ·; V ar(X) = 2´2 :

10
 d
Remark. The notation = means equal in distribution. Without further notice, »i ; »~i ; »;
and »~ denote i.i.d. exponential random variables with mean ´ and variance ´ 2 .
By the memoryless property, we have the conditional distribution
d d
(»1 ¡ »2 j»1 > »2 ) = »; (»1 ¡ »2 j»1 < »2 ) = ¡»;

thus leading to the conclusion that

( )
d »; with probability 1/2
»1 ¡ »2 = ; (3.4)
¡»; with probability 1/2

because the probabilities of the events »1 > »2 and »1 < »2 are equal. Therefore, we have
8 9 8 9
>
> »1 + »2 ; 1/4 >
> >
> »1 + »2 ; 1/4 >
>
>
< >
= >
< >
=
d »1 ¡ »2 ; 1/4 d »1 ; 1/4
X1 + X2 ¡ 2· = = ;
>
> ¡»1 + »2 ; 1/4 >
> >
> ¡»1 ; 1/4 >
>
>
: >
; >
: >
;
¡»1 ¡ »2 ; 1/4 ¡»1 ¡ »2 ; 1/4

in other words,
( P )
M
i=1 »i ;
P
d with probability 1/2
X1 + X2 ¡ 2· = M ; (3.5)
¡ i=1 »i ; with probability 1/2

where »i are i.i.d. exponential random variables, and M is a discrete random variable with
P (M = 1) = 1=2, and P (M = 2) = 1=2: Similar computation leads to
( P )
M
i=1 »i ;
P
d with probability 1/2
X1 + X2 + X3 ¡ 3· = M ; (3.6)
¡ i=1 »i ; with probability 1/2

where M is a discrete random variable with P (M = 3) = 1=4, P (M = 2) = 3=8, and P (M =

1) = 3=8. The following proposition extends (3.5) and (3.6).
Proposition 1. We have that, for every n ¸ 1, the following decomposition
n
( PM(n) )
X
Xi ¡ n· =
d i=1 »i ; with probability 1/2
; (3.7)
P M(n)
i=1 ¡ i=1 »i ; with probability 1/2

where »i are i.i.d. exponential random variables with mean ´, and M(n) is a discrete random
variable, independent of the X's, with
à !
2j 2n ¡ j ¡ 1
P (M(n) = j) = ; 1 · j · n; (3.8)
22n¡1 n¡1
¡0¢
where 0 is de¯ned to be 1.
Proof. See the appendix.

11
 As a key step in deriving closed form solutions, this proposition indicates that the sum of
i.i.d. double exponential random variables can be written, in distribution, as a single (ran-
domly) mixed double gamma random variable. This decomposition simpli¯es the calculation
enormously, because, by conditioning on M(n), we have only one gamma random variable to
deal with. A general result similar to the decomposition (3.7) was ¯rst discovered in Shanthiku-
mar (1985), although (3.8) gives a more explicit result for the distribution of M(n). Further-
more, the proofs are totally di®erent: a combinatorial approach is used here, while the proof in
Shanthikumar (1985) is based on the Laplace transform.

3.3. Hh functions

For option pricing, it is necessary to consider the distribution of the sum of normal and double
exponential (or gamma) random variables. Fortunately, this distribution can be obtained in
closed form, thanks to a special function, Hh function, from the mathematical physics literature.
The de¯nition of the Hh function, for example on p. 691 of Abramowitz and Stegun (1972), is
given by
Z 1 Z 1
1 2 =2
Hhn (x) = Hhn¡1 (y) dy = (t ¡ x)n e¡t dt ¸ 0; n = 0; 1; 2; ::: (3.9)
x n! x

d
Hhn (x) = ¡Hhn¡1 (x) ; n = 0; 1; 2; :::
dx
and
2 =2 p p
Hh¡1 (x) = e¡x = 2¼'(x); Hh0 (x) = 2¼©(¡x):

Therefore, the Hh function can be viewed as a generalization of the cumulative normal distri-
bution function. Notice that, for a standard normal random variable Z,
n!
E ((Z ¡ x)n I (Z ¸ x)) = p Hhn (x) ; n = 0; 1; 2; :::
2¼
Furthermore, for every n ¸ 0, the Hh function is a non-increasing function such that

Hhn (x) ! 0; as x ! 1, n = ¡1; 0; 1; 2; :::;

Hhn (x) ! 1; as x ! ¡1, n = 1; 2; 3; :::; (3.10)

p p
Hh0 (x) = 2¼©(¡x) ! 2¼; as x ! ¡1.

There is also an interesting connection between the Hh function and the parabolic cylinder
function, U , which is widely used in mathematical physics:
1 2
U (n + ; x) = ex =4 Hhn (x): (3.11)
2

12
 Notice that if x < 0 then the integrand in (3.9) has a maxima at
p
x + x2 + 4n
;
2
and most of the value of the integral is realized around the peak area. Therefore, for numerical
computation, we want to split the integral more around the peak area.
The Hh function can be computed, fast and easily, by evaluating the integral in (3.9) directly
using software packages, such as Mathematica, Maple, Splus, or Matlab. Alternatively, it can
also be obtained by using a table in Abramowitz and Stegun (1972) for the parabolic cylinder
functions U , along with (3.11). The following is a mathematica code for the Hh function.
Notice in the code the function is set to be zero if x ¸ 10, because the value is less than 10¡100 .
In addition, we only need more interval spliting if x < ¡6.
Hh[x , n ] := If[x >= -6 , If [x <10,
1/n!*NIntegrate[(t - x)^n *Exp[-t^2/2], ft, x, Infinity g ], 0],
(temp = (x + Sqrt[x*x +4*n])*0.5;
( NIntegrate[(t - x)^n*Exp[-t^2/2], ft, x, temp-3 g]+
NIntegrate[(t - x)^n*Exp[-t^2/2], ft, temp-3 , temp-1 g]+
NIntegrate[(t - x)^n *Exp[-t^2/2], ft, temp-1 , temp g]+
NIntegrate[(t - x)^n*Exp[-t^2/2], ft, temp, temp+1 g]+
NIntegrate[(t - x)^n*Exp[-t^2/2], ft, temp+1 , temp+3 g]+
NIntegrate[(t - x)^n*Exp[-t^2/2], ft, temp+3, Infinity g] ) /n!)]
The following are some pictures for the Hh function. It takes only about 15 seconds to
generate all the pictures on a Pentium 400 by using the software package Maple.

12
80
10

60 8

6
40
4
20
2

-4 -2 00 2 4 -4 -2 00 2 4
x x

Figure 3.3: Hh functions for n = 1; 3; 5. Figure 3.4: n = 1:

13
 50 80

40
60

30
40
20

20
10

-4 -2 00 2 4 -4 -2 00 2 4
x x

Figure 3.5: n = 3. Figure 3.6: n = 5.

The Hh function can be computed very fast. As a hypothetical example, though not relevant
to the ¯nance problems we are studying, it only takes a fractional CPU time on a Pentium 400
to get that Hh[¡1000; 100] = 2:6992 ¤ 10142 . The following proposition, which will be used
later, con¯rms what are observed from the above pictures about the limiting behavior of the
Hh function. More precisely, the left tail of an Hh function has a polynomial growth rate, and
the right tail has an exponential decay rate.
Proposition 2. For every n ¸ ¡1, as x ! 1,
1 2 =2
Hhn (x) » e¡x : (3.12)
xn+1
and as x ! ¡1,
Hhn (x) = O(jxjn ): (3.13)

Proof. See the appendix.

For option pricing it is important to evaluate the integral In ,
Z 1
In = e®x Hhn (¯x ¡ ±) dx; n ¸ 0
¸

for arbitrary constants ®, ¸, and ¯.

Proposition 3. If ¯ > 0, then, for all n ¸ 0,
Z 1
In = e®x Hhn (¯x ¡ ±) dx
¸
n µ ¶n¡i µ ¶n+1 p
e®¸ X ¯ ¯ 2¼ ®± +®
2
®
= ¡ Hhi (¯¸ ¡ ±) + e ¯ 2¯2 ©(¡¯¸ + ± + ): (3.14)
® i=0 ® ® ¯ ¯

If ¯ < 0 and ® < 0, then, for all n ¸ 0,

Z 1
In = e®x Hhn (¯x ¡ ±) dx
¸

14
 n µ ¶n¡i µ ¶n+1 p
e®¸ X ¯ ¯ 2¼ ®± 2
+ ®2 ®
= ¡ Hhi (¯¸ ¡ ±) ¡ e ¯ 2¯ ©(¯¸ ¡ ± ¡ ): (3.15)
® i=0 ® ® ¯ ¯

If ¯ < 0 and ® > 0, then, for all n ¸ 0,

Z 1
In = e®x Hhn (¯x ¡ ±) dx = 1:
¸

Proof. See the appendix.

3.4. Sum of double exponential and the normal random variables

Let X be a double exponential random variable with density function

1 ¡jxj=´
fX (x) = e :
2´
¡ ¢
Let Y be a standard normal distribution N 0; ¾2 . A lengthy, though straightforward, calcu-
lation shows that the density of X + Y is given by
( Ã ! Ã !)
1 2 2 1 ¡t=´ t´ ¡ ¾2 1 t´ + ¾2
fX+Y (t) = e¾ =(2´ ) e © + et=´ © ¡ ; (3.16)
´ 2 ¾´ 2 ¾´
and
µ ¶ Ã ! Ã !
u 1 2 2 u´ ¡ ¾ 2 1 2 2 ¡u´ ¡ ¾2
P (X + Y ¸ u) = © ¡ + e¡u=´ e¾ =(2´ ) © ¡ eu=´ e¾ =(2´ ) © :
¾ 2 ¾´ 2 ¾´

Furthermore, for c < 1=´,

E(aeb+c(X+Y ) ¡ K)+
½ ¾ µ ¶
1 1=2 1=2 h
= ae expf ¾2 c2 g©(¡h=¾ + ¾c)
b
+ ¡ K© ¡
2 1 + c´ 1 ¡ c´ ¾
½ ¾ Ã !
1 ¡h=´ ¾2 =(2´2 ) 1 h´ ¡ ¾2
+ e e K ¡1 ©
2 1 ¡ c´ ¾´
½ ¾ Ã !
1 2 2 1 h´ + ¾2
+ eh=´ e¾ =(2´ ) K 1 ¡ © ¡ :
2 1 + c´ ¾´
More generally, we get the following proposition.
Proposition 4. Let X be a random variable such that
( P )
n
i=1 »i
P
with probability p
X= n ;
¡ i=1 »i with probability 1 ¡ p

with the density function

(1=´)e¡x=´ (x=´)n¡1 (1=´)ex=´ (¡x=´)n¡1
fX (x) = p I (x ¸ 0) + (1 ¡ p) I (x < 0) ;
(n ¡ 1)! (n ¡ 1)!

15
 ¡ ¢
and Y be a normal random variable with distribution N 0; ¾2 . Then, in terms of the Hh
function, the density of X + Y is given, for n ¸ 1, by
2 2
( Ã ! Ã !)
¾n e¾ =(2´ ) ¡t=´ t´ ¡ ¾2 t=´ t´ + ¾ 2
fX+Y (t) = n p pe Hhn¡1 ¡ + (1 ¡ p) e Hhn¡1 ; (3.17)
´ ¾ 2¼ ¾´ ¾´
for n ¸ 1,
n¡1 µ ¶i à !
u 2 2 X ¾ 1 ¡u´ + ¾2
P (X + Y ¸ u) = ©(¡ ) + pe¡u=´ e¾ =(2´ ) p Hhi (3.18)
¾ i=0
´ 2¼ ¾´
à !
Xµ
n¡1
¾
¶i
1 u´ + ¾2
¾ 2 =(2´ 2 )
¡ (1 ¡ p) eu=´ e p Hhi ;
i=0
´ 2¼ ¾´
for n ¸ 1 and c < 1=´,

E(aeb+c(X+Y ) ¡ K)+
µ ¶
b p (1 ¡ p) 2 2 h
= ae n
+ n
ec ¾ =2 ©(¡h=¾ + c¾) ¡ K©(¡ )
(1 ¡ c´) (1 + c´) ¾
à !
Xµ
n¡1
1
¶ µ ¶i
¾ 1 ¡h´ + ¾2
¡h=´ ¾ 2 =(2´ 2 )
+pe e K ¡1 p Hhi (3.19)
i=0
(1 ¡ c´)n¡i ´ 2¼ ¾´
à !
X½
n¡1
1
¾ µ ¶i
¾ 1 h´ + ¾2
h=´ ¾ 2 =(2´ 2 )
+ (1 ¡ p) e e K 1¡ p Hhi ;
i=0
(1 + c´)n¡i ´ 2¼ ¾´
where
log(K=a) ¡ b
h= :
c
Proof. See the appendix.
Remark. As ´ ! 0, which means that the jump sizes are getting smaller and smaller, we
have, thanks to (3.12),
2 2
à !
¾n¡1 e¾ =(2´ ) ¡t=´ t´ ¡ ¾2
p e Hhn¡1 ¡
´n 2¼ ¾´
2 2 µ ¶¡n ( µ ¶2 )
¾n¡1 e¾ =(2´ ) ¡t=´ ¡t ¾ 1 ¡t ¾
» p e + exp ¡ +
´n 2¼ ¾ ´ 2 ¾ ´
µ ¶¡n ( )
1 ¡t´ t2
= p +1 exp ¡ 2
¾ 2¼ ¾2 2¾
( )
1 t2
! p exp ¡ 2 ;
¾ 2¼ 2¾
in other words,
2 2
à !
¾n¡1 e¾ =(2´ ) ¡t=´ t´ ¡ ¾2
p e Hhn¡1 ¡ ! fY (t); as ´ ! 0; (3.20)
´n 2¼ ¾´

16
which is the density function of Y . Similarly, as ´ ! 0,
2 2
à !
¾n¡1 e¾ =(2´ ) t=´ t´ + ¾2
p e Hhn¡1 ! fY (t); as ´ ! 0: (3.21)
´n 2¼ ¾´

Therefore, taking the limit in (3.17) yields

fX+Y (t) ! fY (t); as ´ ! 0;

the limit in (3.18)

u
P (X + Y ¸ u) ! ©(¡ ) = P (Y ¸ u); as ´ ! 0;
¾
via (3.20) and (3.21); and the limit in (3.19)
2 ¾ 2 =2 h
E(aeb+c(X+Y ) ¡ K)+ ! aeb ec ©(¡h=¾ + c¾) ¡ K©(¡ ) = E(aeb+cY ¡ K)+ ;
¾
via (3.20) and (3.21); all as we expect what should be.

4. Option Pricing

4.1. European call and put options

Theorem 1. The price of a European call option in (3.1) is given by

1 X
n
à !
X (¸T )n 2j 2n ¡ j ¡ 1
Ãc (0) = e¡¸T 2n¡1
¢
n=1 j=1
n! 2 n¡1
µ ¶
¡¸³T +n· 1 1 1
¢fS (0) e j
+ ©(a+ ) ¡ e¡rT K©(a¡ )
2 (1 ¡ ´) (1 + ´)j
X j¡1 µ ¶ Ã p !i
1 2 2 1 ¾ T 1
+ e¡rT e¡h=´ e¾ T =(2´ ) K ¡1 p Hhi (c¡ )
2 i=0
(1 ¡ ´)j¡i ´ 2¼
j¡1 ½ ¾ Ã p !i
1 2 2 X 1 ¾ T 1
+ e¡rT eh=´ e¾ T =(2´ ) K 1¡ j¡i
p Hhi (c+ )g
2 i=0
(1 + ´) ´ 2¼
n o
+e¡¸T S(0)e¡¸³T ©(b+ ) ¡ Ke¡rT ©(b¡ ) ;

where
³ 2
´
log(S(0)=K) + r § ¾2 ¡ ¸³ T + n·
a§ = p ;
¾ T
³ ´
¾2
log(S(0)=K) + r § 2 ¡ ¸³ T
b§ = p ;
¾ T

17
 p
¾ T h
c§ = § p ;
´ ¾ T
à !
¾2
h = log(K=S(0)) + ¸³T ¡ r ¡ T ¡ n·;
2
e·
³ = ¡ 1:
1 ¡ ´2
Proof. See the appendix.
Remark. There are two special cases worthy of mentioning.
Case 1. Suppose · = 0. As ´ ! 0, which means that the jump sizes are getting smaller and
smaller, the formula converges to the celebrated Black-Scholes formula. In fact, as ´ ! 0, we
1
have ³ = 1¡´ 2
¡ 1 ! 0, and, by the dominated convergence theorem (the theorem is applicable
because the two terms involving summations over i can be, up to a constant, bounded from
P1 ¡¸T (¸T )
n
above by ´, via (3.20) and (3.21), and we have the fact that n=1 n´e n! = ¸T ´ ! 0),
1 X
n
à !
X (¸T )n 2j 2n ¡ j ¡ 1
Ãc (0) ! e¡¸T 2n¡1
fS (0) ©(a+ ) ¡ e¡rT K©(a¡ )g
n=1 j=1
n! 2 n ¡ 1
n o
+e¡¸T S(0)©(b+ ) ¡ Ke¡rT ©(b¡ )
1 n
à !
X
¡¸T (¸T )n 1 X 2n ¡ j ¡ 1
= e fS (0) ©(a+ ) ¡ e¡rT K©(a¡ )g 2n¡1 2j
n=1
n! 2 j=1
n¡1
n o
+e¡¸T S(0)©(b+ ) ¡ Ke¡rT ©(b¡ )
1
X (¸T )n n o
= e¡¸T fS (0) ©(b0+ ) ¡ e¡rT K©(b0¡ )g + e¡¸T S(0)©(b0+ ) ¡ Ke¡rT ©(b0¡ )
n=1
n!
= S (0) ©(b0+ ) ¡ e¡rT K©(b0¡ );

where we have used the identity

n
à !
X 2n ¡ k ¡ 1
22n¡1 = 2k ;
k=1
n¡1

which will be proved in the appendix at the end of the proof of Proposition 1, and the fact that
in this special case ³ ´
2
log(S(0)=K) + r § ¾2 T
0 p
a§ = b§ = b§ := :
¾ T
Case 2. ¸ = 0, which means that there is no jumps at all. Then

Ãc (0) = S(0)©(b0+ ) ¡ Ke¡rT ©(b0¡ );

18
which is the Black-Scholes formula, where
³ ´
¾2
log(S(0)=K) + r § 2 T
b0§ = p :
¾ T
Remark. About the put-call parity.

P ut ¡ Call = e¡rT E¤ ((K ¡ S(T ))+ ¡ (S(T ) ¡ K)+ )

= e¡rT E¤ (K ¡ S(T ))
= Ke¡rT ¡ S(0);

via (3.2), from which the price of a put option can be obtained from the formula for the call
option.
Remark. About hedging. As we mentioned, the market is basically incomplete, and the
usual hedging arguments are not applicable here. However, as pointed out by Merton (1976),
assuming the jump risks are non-systematic and independent for di®erent stocks, it is still
possible to use the traditional delta hedging to get an asymptotic riskless hedging portfolio, as,
by the law of large numbers, the unhedged risks will cancel each other asymptotically.
Remark. About programming. In mathematica and many other software packages, it is
better to convert the sum into matrix operations. A mathematica code can be downloaded
from the author's web page.
Remark. Although the pricing formulae involve in¯nite series, numerically the expressions
can be evaluated very quickly to a high degree of accuracy through truncation. Our experi-
ence shows that numerically only the ¯rst 10 terms in the in¯nite series are needed for most
applications.

4.2. Pricing of options on futures

Assume, for now, the term structure of interest rate is °at, and r is a constant. Then the
futures price, F (t; T ¤ ), with maturity at time T ¤ , is given by
¤ ¡t)
F (t; T ¤ ) = er(T S(t):

Theorem 2. The price of a European call option on a futures contract is given by

Ãc (D; F (0; T ); K; T; ·; ´; ¸; ¾)
1 X
n
à !
X
¡¸T (¸T )n 2j 2n ¡ j ¡ 1
= D¢( e ¢
n=1 j=1
n! 22n¡1 n¡1

19
 µ ¶
¤ ¡¸³T +n· 1 1 1
¢fF (0; T )e j
+ ©(a+ ) ¡ K©(a¡ )
2 (1 ¡ ´) (1 + ´)j
X j¡1 µ ¶ Ã p !i
1 2 2 1 ¾ T 1
+ e¡h=´ e¾ T =(2´ ) K ¡1 p Hhi (c¡ )
2 i=0
(1 ¡ ´)j¡i ´ 2¼
j¡1 ½ ¾ Ã p !i
1 2 2 X 1 ¾ T 1
+ eh=´ e¾ T =(2´ ) K 1¡ j¡i
p Hhi (c+ )g
2 i=0
(1 + ´) ´ 2¼
n o
+e¡¸T F (0; T ¤ )e¡¸³T ©(b+ ) ¡ K©(b¡ ) );

where
log(F (0; T ¤ )=K) § ¾2 T =2 ¡ ¸³T + n·
a§ = p ;
¾ T
log(F (0; T ¤ )=K) § ¾2 T =2 ¡ ¸³T
b§ = p ;
¾ T
p
¾ T h
c§ = § p ;
´ ¾ T
¾2 T
h = log(K=F (0; T ¤ )) + ¸³T + ¡ n·;
2
e·
³ = ¡1
1 ¡ ´2
D = e¡rT :

This theorem can be proved easily by using Theorem 1 and standard proofs about the
futures options under jump di®usion models (see, for example, Musiela and Rutkowski, 1997,
and Birge and Kou, 1999). Hence the proof is omitted. Notice, however, we cannot simply
prove Theorem 2 by plunging in the futures formula into Theorem 1, as the underlying asset
now is the futures contract rather than the spot asset itself.
Remark. Again for the following two special cases, the formula degenerates to the Black's
futures option formula: case 1, · = 0 and ´ ! 0, which means that the jump sizes are getting
smaller and smaller; and case 2, ¸ = 0, which means that there is no jumps at all. Recall that
the Black's formula is
Ãc (0) = e¡rT fS(0)©(b0+ ) ¡ K©(b0¡ )g;

where
log(F (0; T ¤ )=K) § ¾2 T =2
b0§ = p :
¾ T
Remark. We can also have a similar put-call parity for the futures options:

P ut ¡ Call = e¡rT (K ¡ F (0; T ¤ )):

20
5. Pricing of interest rate derivatives and \volatility smile"

In this section, we ¯rst derive closed form solution for the prices of some interest rate derivatives,
particularly interest rate caps and °oors, in the double exponential jump model of the previous
sections; and then we will use the formulae and a real data set of interest rate caplets to show
that the model is capable to produce \volatility smile".
Interest rates with jumps have been studied by many papers. Equilibrium asset pricing
models for interest rates with jumps include Ahn and Thompson (1988), Attari (1996), Das
and Foresi (1996), Nietert (1997); the pricing of interest rate derivatives in the presence of jumps
is considered in BjÄ
ork, Kabanov , and Runggaldier (1997), Burnetas and Ritchken (1997), Das
and Foresi (1996), Du±e and Kan (1996), Du±e, Pan, and Singleton (1998), Jarrow and Madan
(1995), and Shirakawa (1991). Various sources of jumps in interest rates, including moves by
central banks, are investigated in Babbs and Webber (1997), Das (1998), El-Jahel, Lindberg,
and Perraudin (1997), and Honor¶e (1998). In addition, the possibility of default (as modeled
in Du±e and Singleton, 1996, and Jarrow and Turnbull, 1995) provides further motivation for
including jumps, though we do not consider credit risk here.
In particular, a general framework of pricing interest rate caps and °oors with jump risk has
been studied in detail in Glasserman and Kou (1999) (see also Jamshidian, 1999). The results
in this section is a corollary of that paper, with the logarithm of the jump sizes being specialized
to be double exponentially distributed, except that we have to do the explicit calculation in
the case of the double exponential distribution. Therefore, we only give a sketch of the results,
and refer the readers to that paper for details.

5.1. Pricing interest rate caps and °oors

Interest rate caps and °oors are among the most liquated of all interest rate derivatives. To
study them, it is necessary to consider the term structures based on simple forward rates with
a ¯xed accural period ± expressed as a fraction of a year; for example, ± = 1=4 means three
month rates. With ± ¯xed, we denote by L(t; T ) the forward rate for the interval from T to
T + ± as of time t · T . Thus, a party entering into a contract at time t to borrow $1 over the
interval [T; T + ±] will receive $1 at time T and will pay to the lender $(1 + ±L(t; T )) at time
T + ±. Mathematically speaking, the forward rate is de¯ned as
µ ¶
1 B(t; T )
L(t; T ) = ¡1 ;
± B(t; T + ±)

21
where B(t; ¿ ) is the time-t price of a zero coupon bond maturing at time ¿ . An interest rate
caplet for the period [Tn ; Tn+1 ], with Tn = n±, and struck at K is a derivative security paying
±(Ln (Tn ) ¡ K)+ at time Tn+1 , where Ln (t) = L(t; n±):
Under the marked-point-processes model in Glasserman and Kou (1999), particularly equa-
(i)
tion (9) in that paper with log Xn being double exponential random variables instead of being
normal, we can explicitly ¯nd the time-t price Cn (t) of the nth caplet if (25) in Glasserman and
Kou (1999) holds with fn a double exponential density. In particular, if ¾k are deterministic
then the time-t price of the nth caplet, t < Tn , is given by

Cn (t) = ±Ãc (Bn+1 (t); Ln (t); K; Tn ¡ t; ·n ; ´n ; ¸n ; ¾n ) ; (5.1)

where we have used the same notation as in Theorem 2. Obviously, for Tn · t · Tn+1 ,

Cn (t) = ±Bn+1 (t)(Ln (Tn ) ¡ K)+ :

Here ·n , ´n , ¸n , and ¾n are risk-neutral parameters associated with the nth caplet. In other
words, one can price interest rate caplets by using the formulae for call options on the futures
contracts.
By summing the prices of individual caplets one can price a cap, which is simply a portfolio
of caplets with consecutive maturities. Similarly, one can price an interest rate °oor by summing
the prices of single-period °oors, via the formulae for put options on futures contracts.
Remark. As noted in, for example, Hull (1999), Miltersen, Sandmann, and Sondermann
(1997), Glasserman and Kou (1999), from the prices of caps and °oors it is possible to derive
prices of puts and calls on zero coupon bonds, provided the maturity of the bond is one period
later in the tenor structure than the expiration of the option. In particular, a put option on
the bond struck at K can be valued as a portfolio of K caplets struck at (1 ¡ K)=(±K).

5.2. \Implied volatility smile"

To illustrate that this model can produce \implied volatility smile", we consider a real data
used ¯rst in Andersen and Andreasen (1999) for 2-year and 9-year caplets in the Japanese
LIBOR market as of late May 1998. In this example, we choose a set of model parameters,
calculate caplet prices at a range of strikes using (5.1), and then ¯nd the corresponding implied
volatilities based on the Black's formula. More precisely, these implied volatilities are the values
of ¾T that equate the price computed using the Black's formula to the price computed using
(5.1) with all other parameters held ¯xed. Their implied volatilities based on mid-market prices

22
Figure 5.1: Mid-market and model implied volatilities for Japanese LIBOR caplets in May 1998.
The parameters used for the model implied volatilities are ·4 = ¡0:83, ´4 = 0:35, ¸4 = 0:78,
¾4 = 0:21, and ·18 = ¡0:63, ´18 = 0:45, ¸18 = 0:14, ¾18 = 0:08.

are reproduced in Figure 5.1. Corresponding 6-month forward rates for the same period are
1.181% for the 2-year maturity and 2.913% for the 9-year maturity.1 Figure 5.1 also shows
implied volatility curves derived from (5.1) using the same forward rates and the parameters in
the caption. (With 6-month accrual intervals, the 2-year and 9-year caplets correspond to n = 4
and n = 18 in the notation of (5.1).) The ¯gure suggests the possibility of a very close ¯t even
to a very sharp market skew. Because the parameters used in (5.1) apply under a risk-neutral
measure rather than the physical measure, we can interpret the parameters in Figure 5.1 as
suggesting that the market attaches a large risk premium to the possibility of a sharp downward
movement in Japanese interest rates. A similar interpretation is also given in Glasserman and
Kou (1999) for the normal jump model.
Remark. The good ¯tness of the model to the market data of implied volatilities largely
comes from a large number of free parameters. This is particular useful if the observed implied
volatility smile is truly a \curved smile", which means that it has a right tail going upwards
signi¯cantly rather than being °at as in Figure 5.1. Although the \curved smile" is frequently
1
We thank Leif Andersen of General Re Financial Products for providing these values and also the mid-market
implied volatilities in the ¯gure.

23
observed in markets, it is quite di±cult for other models with fewer parameters, such as the
CEV, to ¯t the curve. Of course, more data would be needed to ¯t the models with more
parameters. So our model is more appropriate for more liquated options. In addition, in actually
trying to ¯t a model to market data, one might also want to impose additional restrictions on
the parameters.
Remark. It should be pointed out that many other models can ¯t the volatility smile well.
For the particular caplet data used here, both the CEV model (see Andersen and Andreasen,
99) and Merton normal jump model (see Glasserman and Kou, 1999) can lead to very similar
¯t. A key di®erence is that the double exponential jump model not only can incorporate
volatility smile, but also has leptokurtic feature and analytical tractability for many other
options, particularly for exotic options.

6. Advantage and disadvantage of the model

Three major advantages of the model are that it can lead to

² leptokurtic feature, namely, compared to the geometric Brownian motion, the model
can produce the desirable features of higher peak and (asymmetric) heavier tails for the
underlying asset;

² \implied volatility smile", which is a puzzle for the Black-Scholes model;

² analytical tractability, namely because of the simple setting of the model, which only
consists of a Brownian part and a Poisson jump part with the logarithm of the jump
sizes being double exponential, it can lead to analytic solutions for various option pricing
problems; in addition, the parameters in the model can be easily interpreted.

Although there are, as we discussed before, many models can incorporate at least one of the
three, leptokurtic feature, \volatility smile", and analytically tractability, the current model
can incorporate all of them under a uni¯ed framework.
The closed form solutions are made possible because of the simplicity of the exponential
distribution and the memoryless property. In a forthcoming paper, closed form solutions for
various exotic options, such as perpetual American options, barrier and lookback options will
be presented. These options cannot be priced in closed form if the logarithm of the jump sizes
have a normal distribution, because of the overshot problem related to the boundary crossing
problems.

24
 One disadvantage is that the formulae, though being analytic, for option pricing appear to
be long. However, this may not a major problem because the Hh function can be computed
easily (in fact a short code would be su±cient in many software packages), and what appears
to be lengthy to human eyes might make little di®erence in terms of computer programming,
as long as it is a closed form solution.
More serious criticisms are (a). whether such a model can be incorporated in an equilibrium
framework; (b) empirical justi¯cation of the model. Hopefully this paper will generate interest
for further researches on both equilibrium and empirical aspects of the model.

25
Appendix: Proofs

To prove Proposition 1, the following lemma is needed.

Lemma 1. Suppose »i and »~i are independent identically distributed exponential random
variables with mean ·. Then
n m
( P ¡ ¢ )
X X d
k
with prob. (1=2)n¡k+m ¢ n+m¡k¡1
i=1 »i ; m¡1 ¢ ; k = 1; :::; n
»i ¡ »~j + (n ¡ m)· = P l n¡l+m
¡n+m¡l¡1 :
i=1 j=1
¡ i=1 »i ; with prob. (1=2) ¢ n¡1 ; l = 1; :::; m
(A.1)
Pn Pm ~
Proof. Introduce the random variables A(n; m) = i=1 »i ¡ j=1 »j + (n ¡ m)·. Then
( ) ( )
d A(n ¡ 1; m ¡ 1) + »n ; 1/2 d A(n; m ¡ 1); 1/2
A(n; m) = = ;
A(n ¡ 1; m ¡ 1) ¡ »n ; 1/2 A(n ¡ 1; m); 1/2
via (3.4). Now image a plane with the horizontal axis representing the number of »i and vertical
axis representing the number of »~j . Suppose we have a random walk on the integer lattice point
of this plane. Starting from any point (n; m), n; m ¸ 1, the random walk goes either one step
to the left or one step down with equal probability 1/2, and the random walk will stop once
it reaches either the horizontal or the vertical axis. For any path that leading point (n; m) to
(k; 0), 1 · k · n, it must reach (k; 1) ¯rst before it makes a ¯nal step to (k; 0). Furthermore, all
the paths going from (n; m) to (k; 1) must have exactly n ¡ k left's and m ¡ 1 down's, whence
¡n¡k+m¡1¢
the total number of such paths is m¡1 . Similarly, the total number of paths that leading
¡n¡l+m¡1¢
point (n; m) to (0; l), 1 · l · m, is n¡1 . Thus,
( P ¡ ¢ )
d
k
i=1 »i ; with prob. (1=2)n¡k+m ¢ n¡k+m¡1
m¡1 ¢ ; k = 1; :::; n
A(n; m) = P l n¡l+m
¡n¡l+m¡1 ;
¡ i=1 »i ; with prob. (1=2) ¢ n¡1 ; l = 1; :::; m
and the lemma is proven. 2
Proof of Proposition 1. It is enough to consider the case · = 0. By the same analogy
used in Lemma 1, to compute probability pk , 1 · k · n, the probability weight assigned to
Pk Pn
i=1 »i when we decompose i=1 Xi , it is equivalent to consider the probability of the random
walk ever reach (k; 0) starting from the diagonal points (i; n ¡ i), 0 · i · n, with probability
¡n¢ 1
of starting from (i; n ¡ i) being i 2n . Notice that the point (k; 0) can only be reached from
points (i; n ¡ i); with the constraint k · i · n ¡ 1, because the random walk stops once it
reaches the horizontal axis. Therefore, for 1 · k · n ¡ 1; (A.1) leads to
n¡1
X
pk = P (going from (i; n ¡ i) to (k; 0)) ¢ P (starting from (i; n ¡ i))
i=k

26
 Xµ
n¡1 ¶i+(n¡i)¡k à !à !
1 i + (n ¡ i) ¡ k ¡ 1 n 1
=
i=k
2 (n ¡ i) ¡ 1 i 2n
Xµ
n¡1 ¶n¡k à !à !
1 n¡k¡1 n 1
=
i=k
2 n¡i¡1 i 2n
Xµ
n¡1 ¶n¡k à !à !
1 n¡k¡1 n 1
= ;
i=k
2 i¡k i 2n
¡0¢
where 0 is de¯ned to be one. Letting j = i ¡ k yields, for 1 · k · n ¡ 1;

X µ
n¡k¡1 ¶n¡k à !à !
1 n¡k¡1 n 1
pk =
j=0
2 j j + k 2n
à !à !
2k n¡k¡1
X n¡k¡1 n
= 2n :
2 j=0
j n¡j¡k

But we have the combinatorial identity

n¡k¡1
à !à ! à !
X n¡k¡1 n 2n ¡ k ¡ 1
= :
j=0
j n¡j ¡k n¡k

Thus, for 1 · k · n ¡ 1, Ã !
2k 2n ¡ k ¡ 1
pk = 2n :
2 n¡k
Of course,
1
pn = :
2n
In summary, we have
à ! à !
2k 2n ¡ k ¡ 1 2k 2n ¡ k ¡ 1
pk = 2n = 2n ; 1·k·n
2 n¡k 2 n¡1
Pk
Similarly, to compute p¡k , k ¸ 1, which is the probability weight assigned to ¡ i=1 »i when
Pn
we decompose i=1 Xi , it is equivalent to consider the probability of the random walk ever
reach (0; k) from the diagonal points (i; n ¡ i), 0 · i · n. This is given by
n¡1
X
p¡k = P (going from (n ¡ i; i) to (0; k)) ¢ P (starting from (n ¡ i; i))
i=k
Xµ
n¡1 ¶(n¡i)+i¡k à !à !
1 (n ¡ i) + i ¡ k ¡ 1 n 1
=
i=k
2 (n ¡ i) ¡ 1 n ¡ i 2n
Xµ
n¡1 ¶n¡k à !à !
1 n¡k¡1 n 1
=
i=k
2 n¡i¡1 i 2n
= pk :

27
Therefore, we can introduce a random variable M(n), such that
à !
2k 2n ¡ k ¡ 1
P (M(n) = k) = 2pk = ; 1 · k · n;
22n¡1 n¡1
and the decomposition is, for n ¸ 2,
n
( PM(n) )
X d
Xi = i=1 »i ; with probability 1/2
;
P M(n)
i=1 ¡ i=1 »i ; with probability 1/2
where »i are i.i.d. exponential random variables. Incidentally, we have also shown that
n
à !
X 2n ¡ k ¡ 1
22n¡1 = 2k ;
k=1
n¡1
P
because (pk + p¡k ) = 1, and thus M(n) is a proper discrete distribution with the domain
f1; 2; :::; ng. 2
Proof of Proposition 2. Case 1. x ! 1. We will prove it by induction. Clearly it is true for
n = ¡1. Now suppose it is true for n = k. Then
1 2
Hhk (x) » k+1 e¡x =2 ;
x
which means that for every ² > 0 it must be that for all large enough x
1 2
1 ¡ ² · Hhk (x)=f k+1 e¡x =2 g · 1 + ²:
x
Therefore, for all large x, we have the following bounds on Hhk+1 (x):
Z 1 Z 1 Z 1
1 ¡y2 =2 1 ¡y2 =2
(1 ¡ ²) k+1
e dy · Hhk+1 (x) = Hhk (y) dy · (1 + ²) k+1
e dy:
x y x x y
But, as x ! 1,
Z 1 µ ¶
1 ¡y2 =2 ¡1¡ 12 k 1 1 2 1 2
k+1
e dy = 2 ¢ ¡ ¡ k; x » k+2 e¡x =2 :
x y 2 2 x
Therefore
1 2
lim Hhk+1 (x)=f k+2 e¡x =2 g = 1;
x!1 x
and the result is proven.
Case 2. x ! ¡1. It is clearly true for n = ¡1. So we only study the situation when n ¸ 0.
As x ! ¡1
Z
1 1 2
Hhn (x) = (t ¡ x)n e¡t =2 dt
n! x
Z
2n 1 n 2
· (jtj + jxjn )e¡t =2 dt
n! x
Z Z
2n 1 n ¡t2 =2 2n 1 2
· jtj e dt + jxjn e¡t =2 dt
n! ¡1 n! ¡1
= O(jxjn );

28
and the proof is terminated. 2
Proof of Proposition 3. We shall study it in serval cases.
Case 1, when ¯ < 0 and ® > 0. Then there is nothing more to say as the integral is 1,
because the integrand goes to in¯nity as x ! 1:
Case 2, when ¯ > 0: In this case, we must have, for n ¸ 0;

e®x Hhn (¯x ¡ ±) ! 0; as x ! 1; (A.2)

for any constant ®, thanks to (3.12). Integration by parts leads to

Z 1
In = e®x Hhn (¯x ¡ ±) dx
¸
Z
1 1
= Hhn (¯x ¡ ±) de®x
® ¸
· Z 1 ¸
1 ®x 1 ®x
= j
Hhn (¯x ¡ ±) e x=¸ ¡ e dHhn (¯x ¡ ±)
® ¸
Z 1
1 ¯
= ¡ Hhn (¯¸ ¡ ±) e®¸ + e®x Hhn¡1 (¯x ¡ ±) dx;
® ® ¸
where the limiting behavior in (A.2) has been applied. In other words, we have a recursion, for
n ¸ 0,
e®¸ ¯
In = ¡ Hhn (¯¸ ¡ ±) + In¡1 ;
® ®
with the initial condition
Z 1
I¡1 = e®x Hh¡1 (¯x ¡ ±) dx
¸
p Z 1 ®x
= 2¼ e '(¡¯x + ±)dx
¸
p
2¼ ®± ®2 ®
= expf + 2 g©(¡¯¸ + ± + ):
¯ ¯ 2¯ ¯
Solving it yields, for n ¸ ¡1,
n µ ¶i µ ¶n+1
e®¸ X ¯ ¯
In = ¡ Hhn¡i (¯¸ ¡ ±) + I¡1
® i=0 ® ®
n µ ¶n¡i µ ¶n+1 p
e®¸ X ¯ ¯ 2¼ ®± ®2 ®
= ¡ Hhi (¯¸ ¡ ±) + expf + 2 g©(¡¯¸ + ± + );
® i=0 ® ® ¯ ¯ 2¯ ¯

where the sum over an empty set is de¯ned to be zero.

Case 3, when ¯ < 0 and ® < 0: In this case, we must also have, for n ¸ 0;

e®x Hhn (¯x ¡ ±) ! 0; as x ! 1; (A.3)

29
for any constant ® < 0, thanks to (3.13). Using integration by parts and (A.3), we again have
a recursion, for n ¸ 0,
e®¸ ¯
In = ¡ Hhn (¯¸ ¡ ±) + In¡1 ;
® ®
but with a di®erent initial condition
Z 1
I¡1 = e®x Hh¡1 (¯x ¡ ±) dx
¸
p Z 1 ®x
= 2¼ e '(¡¯x + ±)dx
¸
p
2¼ ®± ®2 ®
= ¡ expf + 2 g©(¯¸ ¡ ± ¡ ):
¯ ¯ 2¯ ¯
Solving it yields, for n ¸ ¡1,
n µ ¶i µ ¶n+1
e®¸ X ¯ ¯
In = ¡ Hhn¡i (¯¸ ¡ ±) + I¡1
® i=0 ® ®
n µ ¶n¡i µ ¶n+1 p
e®¸ X ¯ ¯ 2¼ ®± ®2 ®
= ¡ Hhi (¯¸ ¡ ±) ¡ expf + 2 g©(¯¸ ¡ ± ¡ );
® i=0 ® ® ¯ ¯ 2¯ ¯
where the sum over an empty set is again de¯ned to be zero. The proof is terminated. 2
Proof of Proposition 4. We prove it in three cases.
Case 1. The density of X + Y . We have

fX+Y (t)
Z 1
= fX (t ¡ x) fY (x) dx
¡1
Z t
(1=´)e¡(t¡x)=´ ((t ¡ x) =´)n¡1 1 2 2
= p p e¡x =(2¾ ) dx
¡1 (n ¡ 1)! ¾ 2¼
Z 1 (1=´)e(t¡x)=´ ((x ¡ t) =´)n¡1 1 2 2
+ (1 ¡ p) p e¡x =(2¾ ) dx
t (n ¡ 1)! ¾ 2¼
Z t ex=´ (t ¡ x)n¡1 1 2 2
= pe¡t=´ (1=´n ) p e¡x =(2¾ ) dx
¡1 (n ¡ 1)! ¾ 2¼
Z
t=´ n
1 e¡x=´ (x ¡ t)n¡1 1 2 2
+ (1 ¡ p) e (1=´ ) p e¡x =(2¾ ) dx
t (n ¡ 1)! ¾ 2¼
Z t (t ¡ x)n¡1 1 2
p e¡(x¡¾ =´) =(2¾ ) dx
2 =(2´ 2 ) 2 2
= pe¡t=´ (1=´n )e¾
¡1 (n ¡ 1)! ¾ 2¼
Z 1 (x ¡ t)n¡1 1 2 2 2
+ (1 ¡ p) e t=´
(1=´ )e n ¾2 =(2´ 2 ) p e¡(x+¾ =´) =(2¾ ) dx:
t (n ¡ 1)! ¾ 2¼
¡ ¢ ¡ ¢
By letting y = x ¡ ¾2 =´ =¾, and y~ = x + ¾2 =´ =¾, we have

fX+Y (t)

30
 Z ¡ ¢n¡1
¡t=´ ¾ 2 =(2´ 2 ) n
t=¾¡¾=´ t ¡ ¾y ¡ ¾2 =´ 1 2
= pe e (1=´ ) p e¡y =2 dy
¡1 (n ¡ 1)! 2¼
Z ¡ ¢n¡1
t=´ ¾2 =(2´ 2 ) n
1 y ¡ ¾2 =´ ¡ t
¾~ 1 2
+ (1 ¡ p) e e (1=´ ) p e¡~y =2 dy
t=¾+¾=´ (n ¡ 1)! 2¼
Z
2 =(2´ 2 )
t=¾¡¾=´ (t=¾ ¡ y ¡ ¾=´)n¡1 1 ¡y2 =2
= pe¡t=´ e¾ ¾n¡1 =´n p e dy
¡1 (n ¡ 1)! 2¼
Z
2 =(2´ 2 )
1 y ¡ ¾=´ ¡ t=¾)n¡1 1 ¡~y2 =2
(~
+ (1 ¡ p) et=´ e¾ ¾n¡1 =´n p e dy
t=¾+¾=´ (n ¡ 1)! 2¼
2 2
e¾ =(2´ ) n o
= p (¾n¡1 =´n ) pe¡t=´ Hhn¡1 (¡t=¾ + ¾=´) + (1 ¡ p) et=´ Hhn¡1 (t=¾ + ¾=´) ;
2¼
because
Z 2 Z 2
1 a e¡y =2 1 1 e¡y1 =2
(a ¡ y)n¡1 p dy = (a + y1 )n¡1 p dy1
(n ¡ 1)! ¡1 2¼ (n ¡ 1)! ¡a 2¼
1
= p Hhn¡1 (¡a) ;
2¼
and similarly Z
1 1 1 2 1
(y ¡ b)n¡1 p e¡y =2 dy = p Hhn¡1 (b) ;
(n ¡ 1)! b 2¼ 2¼
where y1 = ¡y; and the proof of Case 1 is terminated.
Case 2. P (X + Y ¸ u). We have, via the density given in (3.17), that
Z 1
P (X + Y ¸ u) = fX+Y (t) dt
u
2 2
( Z Ã ! Z Ã ! )
¾n e¾ =(2´ ) 1
¡t=´ t´ ¡ ¾2 1
t=´ t´ + ¾2
= p p e Hhn¡1 ¡ dt + (1 ¡ p) e Hhn¡1 dt :
´ n ¾ 2¼ u ¾´ u ¾´
Substituting ¯ = (¡1=¾) < 0, ® = (¡1=´) < 0, ± = (¡¾=´); and ¸ = u in (3.15) yields
Z Ã !
1
¡t=´ t´ ¡ ¾2
e Hhn¡1 ¡ dt
u ¾´
n¡1
X
= ´e¡u=´ (´=¾)n¡1¡i Hhi (¡u=¾ + ¾=´)
i=0
p ¾2
+ (´=¾)n ¾ 2¼ expf¡¾ 2 =´ 2 + 2 g©(¡u=¾ + ¾=´ ¡ ¾=´)
2´
n¡1
X p ¾2
= ´e¡u=´ (´=¾)n¡1¡i Hhi (¡u=¾ + ¾=´) + (´=¾)n ¾ 2¼ expf¡ 2 g©(¡u=¾);
i=0
2´

and substituting ¯ = (1=¾) > 0, ® = (1=´), ± = (¡¾=´); and ¸ = u in (3.14) yields

Z Ã !
1
t=´ t´ + ¾2
e Hhn¡1 dt
u ¾´

31
 Xµ
n¡1
´
¶n¡1¡i
u=´
= ¡´e Hhi (u=¾ + ¾=´)
i=0
¾
µ ¶n p
´ ¾2
+ ¾ 2¼ expf¡¾ 2 =´2 +
g©(¡u=¾ ¡ ¾=´ + ¾=´)
¾ 2´2
X µ ´ ¶n¡1¡i
n¡1 µ ¶n p
´ ¾2
= (¡´)eu=´ Hhi (u=¾ + ¾=´) + ¾ 2¼ expf¡ 2 g©(¡u=¾):
i=0
¾ ¾ 2´

Therefore,

P (X + Y ¸ u)
n¡1
" µ ¶n¡1¡i à ! µ ¶n p #
¾n 1 X ´ ¡u´ + ¾2 ´
¾ 2 =(2´ 2 ) ¡u=´ ¡¾ 2 =(2´ 2 )
= p e fp ´e Hhi + ¾ 2¼e ©(¡u=¾)
´ n ¾ 2¼ i=0
¾ ¾´ ¾
" Ã ! #
Xµ
n¡1
´
¶n¡1¡i
u´ + ¾ 2
µ ¶n p
´
u=´ ¡¾ 2 =(2´ 2 )
+ (1 ¡ p) ¡´e Hhi + ¾ 2¼e ©(¡u=¾) g
i=0
¾ ¾´ ¾
" Ã ! #
Xµ
n¡1
´
¶¡i
1 ¡u´ + ¾2 u
¾ 2 =(2´2 ) ¡u=´ p Hhi ¡¾ 2 =(2´ 2 )
= e fp e +e ©(¡ )
i=0
¾ 2¼ ¾´ ¾
" Ã ! #
Xµ
n¡1
´
¶¡i
1 u´ + ¾2 u
u=´ p Hhi ¡¾ 2 =(2´2 )
+ (1 ¡ p) ¡e +e ©(¡ ) g;
i=0
¾ 2¼ ¾´ ¾

and the proof of Case 2 is terminated.

Case 3. E(aeb+c(X+Y ) ¡ K)+ . By using the density given in (3.17), we get, for n ¸ 1

E(aeb+c(X+Y ) ¡ K)+
Z 1
= (aeb+ct ¡ K)fX+Y (t)dt
h
Z 1
= aeb ect fX+Y (t)dt ¡ KP (X + Y ¸ h)
h
Z Ã !
¾n 1 2 2
1 t´ ¡ ¾2
= aeb n p e¾ =(2´ ) fp ct ¡t=´
e e Hhn¡1 ¡ du
´ ¾ 2¼ h ¾´
Z Ã !
1
ct t=´ t´ + ¾2
+ (1 ¡ p) e e Hhn¡1 dug
h ¾´
n¡1 µ ¶i à !
h 2 2 X ¾ 1 ¡h´ + ¾2
¡K©(¡ ) ¡ Kpe¡h=´ e¾ =(2´ ) p Hhi
¾ i=0
´ 2¼ ¾´
à !
Xµ
n¡1
¾
¶i
1 h´ + ¾2
h=´ ¾2 =(2´ 2 )
+K (1 ¡ p) e e p Hhi ;
i=0
´ 2¼ ¾´

via (3.18), where

log(K=a) ¡ b
h= :
c

32
Substituting ¯ = (¡1=¾) < 0, ® = (c ¡ 1=´) < 0, ± = (¡¾=´); and ¸ = h in (3.15)
Z Ã !
1 t´ ¡ ¾2
ect e¡t=´ Hhn¡1 ¡ dt
h ¾´
Xµ
e(c¡1=´)h n¡1 1
¶n¡1¡i
= ¡ ¡ Hhi (¡h=¾ + ¾=´)
c ¡ 1=´ i=0 (c ¡ 1=´)¾
µ ¶n p
¡1 (c ¡ 1=´)2 ¾2
¡ 2¼(¡¾) expf(c ¡ 1=´)¾2 =´ + g©(¡h=¾ + ¾=´ + (c ¡ 1=´)¾)
(c ¡ 1=´)¾ 2
à !
Xµ
´e(c¡1=´)h n¡1 ´
¶n¡1¡i
¡h´ + ¾ 2
= Hhi
1 ¡ c´ i=0 (1 ¡ c´)¾ ¾´
µ ¶n p
´ 1 1
+¾ 2¼ expf c2 ¾2 ¡ ¾2 =´2 g©(¡h=¾ + c¾);
(1 ¡ c´)¾ 2 2
and substituting ¯ = 1=¾ > 0, ® = (c + 1=´), ± = (¡¾=´); and ¸ = h in (3.14)
Z Ã !
1
ct t=´ t´ + ¾2
e e Hhn¡1 dt
h ¾´
Xµ
e(c+1=´)h n¡1 1
¶n¡1¡i
= ¡ Hhi (h=¾ + ¾=´)
c + 1=´ i=0 (c + 1=´)¾
µ ¶n p
1 (c + 1=´)2 ¾ 2
+ ¾ 2¼ expf(c + 1=´)(¡¾=´)¾ + g©(¡h=¾ ¡ ¾=´ + (c + 1=´)¾)
(c + 1=´)¾ 2
à !
Xµ
´e(c+1=´)h n¡1 ´
¶n¡1¡i
h´ + ¾ 2
= ¡ Hhi
1 + c´ i=0 (c´ + 1)¾ ¾´
µ ¶n p
´ 1 1
+ ¾ 2¼ expf c2 ¾ 2 ¡ ¾ 2 =´2 g©(¡h=¾ + c¾):
(c´ + 1)¾ 2 2
Therefore,

E(aeb+c(X+Y ) ¡ K)+
µ ¶ µ ¶ Ã !
1 2 2 e(c¡1=´)h n¡1
X 1 n¡1¡i
´ ¡i ¡h´ + ¾ 2
= ae p e¾ =(2´ ) fp
b
Hhi
2¼ 1 ¡ c´ i=0 (1 ¡ c´) ¾ ¾´
p
p 2¼ 1 1
+ n
expf c2 ¾2 ¡ ¾2 =´2 g©(¡h=¾ + c¾)
(1 ¡ c´) 2 2
µ ¶ µ ¶ Ã !
´e(c+1=´)h n¡1
X 1 n¡1¡i
´ ¡i h´ + ¾2
¡ (1 ¡ p) Hhi
1 + c´ i=0 (c´ + 1) ¾ ¾´
p
(1 ¡ p) 2¼ 1 1
+ expf c2 ¾2 ¡ ¾ 2 =´ 2 g©(¡h=¾ + c¾)g
(c´ + 1)n 2 2
n¡1 µ ¶i à !
h 2 2 X ¾ 1 ¡h´ + ¾ 2
¡K©(¡ ) ¡ Kpe¡h=´ e¾ =(2´ ) p Hhi
¾ i=0
´ 2¼ ¾´

33
 Ã !
Xµ
n¡1
¾
¶i
1 h´ + ¾2
h=´ ¾ 2 =(2´ 2 ) p Hhi
+K (1 ¡ p) e e :
i=0
´ 2¼ ¾´
After some algebra, we get

E(aeb+c(X+Y ) ¡ K)+
n¡1
à !µ ¶ à !
X aeb+ch i
¾ 1 ¡h´ + ¾2
¡h=´ ¾2 =(2´ 2 )
= pe e ¡K p Hhi
i=0
(1 ¡ c´)n¡i ´ 2¼ ¾´
µ ¶
h p (1 ¡ p) 2 2
¡K©(¡ ) + aeb n
+ n
ec ¾ =2 ©(¡h=¾ + c¾)
¾ (1 ¡ c´) (1 + c´)
n¡1
( )µ ¶ Ã !
X aeb+ch i
¾ 1 h´ + ¾2
h=´ ¾2 =(2´ 2 )
+ (1 ¡ p) e e ¡ +K p Hhi :
i=0
(c´ + 1)n¡i ´ 2¼ ¾´

Since eb+ch = (K=a), we get

E(aeb+c(X+Y ) ¡ K)+
à !
Xµ
n¡1
1
¶ µ ¶i
¾ 1 ¡h´ + ¾2
¡h=´ ¾ 2 =(2´ 2 )
= pe e K ¡1 p Hhi
i=0
(1 ¡ c´)n¡i ´ 2¼ ¾´
µ ¶
p (1 ¡ p) 2 2 h
+aeb + ec ¾ =2 ©(¡h=¾ + c¾) ¡ K©(¡ )
(1 ¡ c´)n (1 + c´)n ¾
à !
X½
n¡1
1
¾ µ ¶i
¾ 1 h´ + ¾2
¾ 2 =(2´2 )
+ (1 ¡ p) eh=´ e K 1¡ p Hhi :
i=0
(1 + c´)n¡i ´ 2¼ ¾´
and the proof is terminated. 2
Proof of Theorem 1. Since N (t) has a Poisson distribution with mean ¸t, conditioning on
N (t), we have
0 0 (Ã ! ) N(T ) 1+ 1
1
X p Y
¾2
Ãc (0) = E¤ @e¡rT @S (0) exp r¡ ¡ ¸³ T + ¾ T Z Vj ¡ K A jN (T ) = nA ¢
n=0
2 j=1
¢P (N (T ) = n)
0 (Ã ! ) 1+
1 n
X (¸T )n ¤ @ ¾2 p Y
= e¡rT e¡¸T E S (0) e¡¸³T exp r¡ T + ¾ TZ Vj ¡ K A
n=0
n! 2 j=1
0 8Ã ! 9 1+
1
X (¸T ) ¤ @n < ¾2 p n
X =
= e¡rT e¡¸T E S (0) e¡¸³T exp r¡ T + ¾ TZ + Xj ¡ K A
n! : 2 ;
n=0 j=1
Q
where recalling that Vj = eXi and over the empty set is set to be one.
By the decomposition (3.7), we have, for n ¸ 1, the following decomposition
n M(n)
X d X
Xi = n· + Yi ;
i=1 i=1

34
where Yi are i.i.d. double random variables
( )
»i ; with probability 1/2
Yi =
¡»i ; with probability 1/2

»i being exponential random variable with mean ´, and M(n) is a discrete random variable
with à !
2j 2n ¡ j ¡ 1
P (M(n) = j) = ; 1 · j · n:
22n¡1 n¡1
Notice that M(n) is independent of all other random variables.
Therefore, conditioning on M(n),

Ãc (0)
0 8Ã ! 9 1+
1
X (¸T )n ¤ @ < ¾2 p M(n) =
X
= e¡rT e¡¸T E S (0) e¡¸³T exp r¡ T + ¾ T Z + n· + Yi ¡ K A
n! : 2 ;
n=0 i=1
1 X
n
à !
¡rT
X
¡¸T (¸T )n 2j 2n ¡ j ¡ 1
= e e 2n¡1
¢
n=1 j=1
n! 2 n¡1
0 8Ã ! 9 1+
< ¾2 p j
X =
¢E¤ @S (0) e¡¸³T exp r¡ T + ¾ T Z + n· + Yi ¡ K A
: 2 ;
i=1
0 Ã (Ã ! ) !+ 1
¾2 p
+e¡rT e¡¸T E¤ @e¡rT S (0) e¡¸³T exp r¡ T + ¾ TZ ¡ K A:
2
³ ´
¾2
Now substituting a = S (0) e¡¸³T , b = r ¡ 2 T + n·, p = 1=2, and c = 1 into (3.19), we have
for 0 < ´ < 1,
0 8Ã ! 9 1+
< ¾2 p j
X =
E¤ @S (0) e¡¸³T exp r¡ T + ¾ T Z + n· + Yi ¡ K A
: 2 ;
i=1
³ 2
´ µ ¶
¡¸³T + r¡ ¾2 T +n· 1 1 1 2 T =2 p p h
= S (0) e j
+ e¾ ©(¡h=(¾ T ) + ¾ T ) ¡ K©(¡ p )
2 (1 ¡ ´) (1 + ´)j ¾ T
X j¡1 µ ¶ à p !i à !
1 2 2 1 ¾ T 1 ¡h´ + ¾2 T
+ e¡h=´ e¾ T =(2´ ) K ¡1 p Hhi p
2 i=0
(1 ¡ ´)j¡i ´ 2¼ (¾ T )´
j¡1 ½ ¾ à p !i à !
1 h=´ ¾2 T =(2´2 ) X 1 ¾ T 1 h´ + ¾2 T
+ e e K 1¡ p Hhi p ;
2 i=0
(1 + ´)j¡i ´ 2¼ (¾ T )´

with à !
log(K=a) ¡ b ¾2
h= = log(K=S(0)) + ¸³T ¡ r ¡ T ¡ n·:
c 2

35
Furthermore, it is easy to check that
à (à ! ) !+
¤ ¾2 p
E S (0) exp r¡ ¡ ¸³ T + ¾ T Z ¡K = S(0)e(r¡¸³)T ©(b+ ) ¡ K©(b¡ );
2

where ³ ´
¾2
log(S(0)=K) + r § 2 ¡ ¸³ T
b§ = p :
¾ T
Thus, we have proved the theorem. 2

References
[1] Abramowitz, M. and Stegun, I.A. (1972). Handbook of Mathematical Function. U. S. Na-
tional Bureau of Standards, 10th Printing.
[2] Ahn, C.M., and Thompson, H.E. (1988) Jump-Di®usion Processes and the Term Structure
of Interest Rates, Journal of Finance, 43, 155-174.
[3] Andersen, T. (1996). Return volatility and trading volume. An information °ow interpre-
tation of stochastic volatility. Journal of Finance. Vol. 51, 169-204.
[4] Andersen, L. and Andreasen, J. (1999). Jump-di®usion processes: volatility smile ¯tting
and numerical methods for pricing. Working paper. General Re Financial Products, New
York.
[5] Attari, M. (1996). Discontinuous Interest Rate Processes: An Equilibrium Model for Bond
Option Prices, working paper, University of Iowa, Iowa City.
[6] Babbs, S., and Webber, N. (1997). Term Structure Modelling under Alternative O±cial
Regimes, in Mathematics of Derivative Securities, M.A.H. Dempster and S.R. Pliska, eds.,
Cambridge University Press.
[7] Barndor®-Nielsen (1998). Probability and statistics: self-decomposability, ¯nance and tur-
bulence. In Probability Towards 2000, edited by L. Accardi and C.C. Heyde. Springer,
NY.
[8] BjÄork, T., Kabanov, Y., and Runggaldier, W. (1997). Bond Market Structure in the Pres-
ence of Marked Point Processes, Mathematical Finance, 7, 211-239.
[9] Black, F. (1976). The pricing of commodity contract. J. of Financial Econ., Vol. 3, 167-179.
[10] Black, F. and Scholes, M. (1973). The pricing of options and corporate liabilities. J. Polit.
Econ., 81 , 637-659.
[11] Blattberg, R.C. and Gonedes, N. (1974). A comparison of the stable and student distribu-
tions as statistical models for stock prices. Journal of Business. Vol 47, 244-280.
[12] Birge, J.R. and Kou, S.G. (1999). Pricing of electricity options under a jump di®usion
model. Working paper, Columbia University.
[13] Boyle, P., Broadie, M., and Glasserman, P. (1997). Simulation methods for security pricing.
Journal of Economic Dynamics and Control. Vol 21, 1267-1321.

36
[14] Brace, A., Gatarek, D., and Musiela, M. (1997). The market model of interest rate dy-
namics. Math. Finance, Vol. 7, 127-155.
[15] Burnetas, A., and Ritchken, P. (1997). On Rational Jump Di®usion Models: An Approach
Using Potentials, Review of Derivatives Research, 1, 325-349.
[16] Clark, P. K. (1973). A subordinated stochastic process model with ¯nite variance for
speculative prices. Econometrica. Vol. 41, 135-155.
[17] Cox, J.T., Ingersoll, J.E., and Ross, S.A. (1985). A theory of the term structure of interest
rates. Econometrica. Vol 53, 385-407.
[18] Cox, J.T. and Rubinstein, M. (1985). Option Markets. Prentice Hall, Englewood Cli®s,
N.J.
[19] Cox, J.T. and Ross, S.A. (1976). The valuation of options for alternative stochastic pro-
cesses. Journal of Financial Economics. Vol 3, 145-166.
[20] Das, S.R. (1998). Poisson-Gaussian Processes and the Bond Markets, NBER Working
Paper No. 6631.
[21] Das, S.R., and Foresi, S. (1996). Exact Solutions for Bond and Option Prices with Sys-
tematic Jump Risk, Review of Derivatives Research, 1, 7-24.
[22] Davydov, D., and Linetsky (1999). The valuation and hedging aof barrier and lookback
options for alternative stochastic processes. Working paper, University of Michigan.
[23] Derman, E. and Kani, I. Riding on a smile. (1994). RISK. Feb. 32-39.
[24] Du±e, D. (1995). Dynamic Asset Pricing Theory. 2nd edn. Princeton University Press,
Princeton.
[25] Du±e, D., and Kan, R. (1996). A Yield-Factor Model of Interest Rates, Mathematical
Finance, 6, 379-406.
[26] Du±e, D., Pan, J., and Singleton, K. (1998). Transform Analysis and Option Pricing for
A±ne Jump-Di®usions, working paper, Graduate School of Business, Stanford University.
[27] Du±e, D., and Singleton, K. (1996). Modeling Term Structures of Defaultable Bonds,
Working paper, Graduate School of Business, Stanford University.
[28] Dupire, B. (1994). Pricing with a smile. RISK. Feb. 18-20.
[29] Eberlein, E. and Keller, U. (1995). Hyperbolic Distributions in Finance. Bernoulli. Vol 1,
281-299.
[30] El-Jahel, L., Lindberg, H., and Perraudin, W. (1997). Interest Rate Distributions,
Yield Curve Modelling and Monetary Policy, in Mathematics of Derivative Securities,
M.A.H. Dempster and S.R. Pliska, eds., Cambridge University Press.
[31] Elliot, R.J. and Kopp, P.E. (1998). Mathematics of Financial Markets. Spinger, New York.
[32] Engle, R. (1982). Autoregressive conditional heteroscedasticity with estimates of the vari-
ance of U.K. in°ation. Econometrica. Vol. 50. 987-1008.

37
[33] Engle, R. (1995). ARCH, Selected Readings. Oxford, Oxford University Press.
[34] Fama, E. (1963). Mandelbrot and the stable Peritian hypothesis. Journal of Business, Vol
36, 297-306.
[35] Fama, E. (1965). The behavior of stock-market prices. Journal of Business, Vol 38, 34-105.
[36] Geman, H., Madan, D., and Yor, M. (1998). Asset prices in Brownian motion: only in
business times. Working paper, University of Paris IX, Dauphine.
[37] Glasserman, P. and Kou, S.G. (1999). The term structure of simple interest rates with
jump risk. Working paper, Columbia University.
[38] Gouri¶eroux, C. (1997). ARCH Models and Financial Applications. Spinger, New York.
[39] Heath, D., Jarrow, R. and Morton, A. (1992). Bond pricing and the term structure models:
a new methodology for contingent claims valuation. Econometrica. Vol 60, 77-105.
[40] Heyde, C.C. (1999). A risky asset model with strong dependence through fractal activity
time. Dept. of Statistics, Columbia University.
[41] Honor¶e, P. (1998). Modelling Interest Rate Dynamics in a Corridor with Jump Processes,
working paper, Department of Finance, Aarhus School of Business, Denmark.
[42] Hull, J.C. (1999). Options, Futures, and Other Derivative Securities, 4th edition, Prentice
Hall, New Jersey.
[43] Hull, J.C. and White, A. (1987). The pricing of options on assets with stochastic volatilities.
Journal of Finance. Vol 42, 281-300.
[44] Hurst, S.R. and Platen, E., and Rachev, S.R. (1997). Subordinated Markov models: a
comparison. Financial Engineering and Japanese Markets. Vol. 4, 97-124.
[45] Ingersoll, J.E. (1988). Theory of Financial Decision Making. Rowman & Little¯eld, Mary-
land.
[46] Jamshidian, F. (1997). LIBOR and swap market models and measures. Finance and
Stochastic. Vol 1. 293-330.
[47] Jamshidian, F. (1999). LIBOR market model with semimartingales. Working paper, Net-
Analytic Limited.
[48] Jarrow, R., and Turnbull, S. (1995). Pricing Options on Financial Securities Subject to
Default Risk, Journal of Finance, 50, 53-86.
[49] Jarrow, R., and Madan, D. (1995). Option Pricing using the Term Structure of Interest
Rates to Hedge Systematic Discontinuities in Asset Returns, Mathematical Finance, 5,
311-336.
[50] Johnson, N., Kotz, S., and Balakrishnan, N. (1995). Continuous Univariate Distribution,
Vol. 2, 2nd ed., Wiley, New York.
[51] Karatzas, I. and Shreve, S. (1998). Methods of Mathematical Finance. Springer, New York.

38
[52] Laplace, P.S. (1774). M¶emoire sur la probabilit¶e des causes par les µev¶enemens, M¶emoires
de Math¶ematique et de Physique, Vol. 6, 621-656.
[53] Madan, D.B. and Seneta, E. (1990). The variance gamma (V.G.) model for share market
returns. Journal of Business, Vol 63, 511-524.
[54] Mandelbrot, B. (1963). The variation of certain speculative prices, Journal of Business,
Vol 36, 394-419.
[55] Mandelbrot, B. (1967). The variation of some other speculative prices, Journal of Business,
Vol 40, 511-524.
[56] Mandelbrot, B., Fisher, A. and Calvet, L. (1997). A multifractal model of asset returns.
Cowles Foundation Discussion Paper No. 1164.
[57] Merton, R. C. (1976). Option pricing when underlying stock returns are discontinuous. J.
of Financial Economics. Vol. 3, 125-144.
[58] Merton, R.C. (1990). Continuous-Time Finance. Blackwell, Cambridge.
[59] Miltersen, K., Sandmann, K., and Sondermann, D. (1997). Closed form solutions for term
structure derivatives with lognormal interest rates. J. of Finance. Vol. 70, 409-430.
[60] Musiela and Rutkowski (1997). Martingale Methods in Financial Modeling. Springer, New
York.
[61] Nietert, B. (1997). Jump/Di®usion Option Pricing | a Reexamination from an Economic
Viewpoint, Department of Finance, Passau University, Germany.
[62] Peters, E.E. (1991). Chaos and Order in the Capital Markets. Wiley, New York.
[63] Peters, E.E. (1994). Fractal Market Analysis. Wiley, New York.
[64] Praetz, P.D. (1972). The distribution of share price changes. Journal of Business. Vol 45,
49-55.
[65] Rogers, L.C.G. (1997). Arbitrage from fractional Brownian motion. Mathematical Finance.
Vol. 7, 95-105.
[66] Rubinstein, M. (1994). Implied binomial trees. Journal of Finance. Vol 49, 771-818.
[67] Samorodnitsky, G. and Taqqu, M.S. (1994). Stable Non-Gaussian Random Processes:
Stochastic Models with In¯nite Variance. Chapman and Hall, New York.
[68] Shanthikumar, J.G. (1985). Bilateral Phase-Type Distributions. Naval Research Logistics
Quarterly. Vol. 32, 119-136.
[69] Shirakawa, H. (1991) Interest Rate Option Pricing with Poisson-Gaussian Forward Rate
Curve Processes, Math. Finance, 1, 77-94.
[70] White, H. (1980). A heteroscedasticity consistent covariance matrix estimation and a direct
test for heteroscedasticy. Econometrica. Vol. 48, 817-838.
[71] Willinger, W. Taqqu, M. and Teverovsky, V. (1999). Stock market prices and long-range
dependence. Finance and Stochastics. Vol 3, 1-13.

39